Resolve types in data generation about compiler HOT 8 CLOSED

To resolve types in Java, we would parse all source (and class/jar) files in the project instead of single files.

How are we going about the language versions of files?

I am thinking of:

• one pass of parsing single files to get language version
• another pass using the highest version to parse all files to build Boa ast with types

from compiler.

There isn't a way to avoid parsing every file twice? Why does the language version matter for this task?

from compiler.

If we tried to parse every file once then it matter because we will only have language version for the entire project.

If we parse twice then each parsing will do a separate task.

I cannot think of anyway to avoid parsing twice.

from compiler.

But why do we need the language version for the project? In reality, a single repo might actually contain multiple language versions.

from compiler.

No, I didn't mean that.
I just meant that we should not use language version that works for all files as the language version of every single file.

from compiler.

Since we are not resolving types for all snapshots, would it be better to store a head snapshot with types separately?

from compiler.

Code
https://github.com/boalang/compiler/blob/datagen2017/src/java/boa/datagen/scm/AbstractConnector.java#L68

Commits
ca964b7
e7b3c5d
c727ede
e3c10c2
e9c490b
5153c47

from compiler.

Use in-project jar files to resolve types
ef1c86a
99e08f8

from compiler.